摘要:
在盲信道均衡或盲语音去混响应用中,盲多信道系统辨识通常是信号解卷积的前提条件,即盲辨识过程后跟一个解卷积过程。本文提出一种基于卡尔曼滤波的同步盲系统辨识与解卷积方法,其中卡尔曼滤波的状态矢量由多信道系统参数与源信号矢量组成,过程方程和测量方程则建立在单输入-多输出系统(SIMO)的输入输出关系及信道间交叉关联关系(Cross Relation)基础上。此外,盲系统辨识部分与解卷积部分是可以解耦的,生成两个看似独立的卡尔曼滤波问题,并且这两个卡尔曼滤波问题可以实现并行计算。与级联结构相比,这种并行结构更有利于算法优化和实时信号处理。仿真表明,对于无噪声理想信号模型,本算法可以实现完全系统辨识和解卷积(信号误差比可达到100 dB以上),说明理论正确;对于实测的混响语音信号亦可以实现一定的去混响效果。
Abstract:
Generally, blind multichannel identification is the prerequisite step for the blind deconvolution of received signals in the field of channel equalization and speech dereverberation. The blind deconvolution procedure is that of blind multichannel identification followed by signal deconvolution. A novel method, which blind multichannel identification and deconvolution are performed synchronically with Kalman filter, is proposed. The state vector is composed of the multichannel impulse responses and the source signal vector, the process and measurement equations are constructed with the Cross Relation conditions and the Input-Output relation of SIMO system. In addition, the blind multichannel identification and the deconvolution parts can be decoupled to generate two seemingly indenpendent Kalman filters, furthermore, the two Kalman filters can be implemented in parallel. Comparing with cascade structure, the parallel structure can be implemented more efficiently and is beneficial to on-line signal processing. Simulations show that the algorithm works perfectly on ideal signal model (the Signal-to-Error Ratio of the deconvolved signal >100 dB) and it also works well for real-word recorded signals.
本站查看
下载:
